NVIDIA's cranked up the virtual workstation caper by giving the world a new GPU that slots into blade servers, plus software to let it run multiple workstation-grade VMs.
The new GPU is the TESLA P6 and uses NVIDIA's Pascal architecture, the company's current flagship. The P6 has 2,048 CUDA cores, 16 GB of memory and uses the MXM form factor so it can slot into blades.
The GPU is offered to those who want to build very dense GPU-enhanced compute rigs for the usual suspects: very graphic applications, machine learning and so on. But NVIDIA's also decided it and the P4, P40 and P100 GPUs should also be put in harness to run virtual workstations.
NVIDIA already offers its “Grid” software for everyday VDI and that code can let multiple desktops share a GPU, be it a Pascal-powered peripheral or a device using the older Maxwell architecture.Theres also Grid for workstations, but that's now been replaced by the new “Quadro Virtual Data Center Workstation Software” that is far more comfortable letting multiple workstations share one GPU.
Among the tricks the new software performs to make this possible is ensuring encoding of graphics is done on the GPU, not the CPU, so that the two work in harmony to make virtual workstations speedy. NVIDIA's enabled this feature on Linux, too. There's also management software that tracks virtual workstation performance so that each VM can be assigned resources appropriate to the workloads users run, instead of guesstimating what they need.
NVIDIA reckons there's a decent market for these virtual things and says a Japanese auto-maker is moving 10,000 engineers to virtual workstations and Pascal GPUs. Microsoft's thinking differently, as it last week announced a new version of Windows 10 just for physical workstations. HP, meanwhile, recently refreshed its physical workstation range and added a backpack workstation for exploring virtual reality. ®